NexusGuard

NexusGuard provides multi-source, pattern-based person tracking for public security and safety operations. It combines computer vision and agentic AI with the ISO/IEC 23090-8 (MPEG Media Orchestration) standard to coordinate streams from heterogeneous cameras and sensors, enabling person re-identification, behavioural-pattern analysis and timely event detection across complex public environments.

Built on VTG’s research lineage in security-domain AI and standardised media-orchestration approaches, NexusGuard integrates with NexusFlow for cross-source agentic reasoning and NexusTrust for EU AI Act-aligned audit trails — ensuring that operational effectiveness and governance accountability are never in conflict.

---

Key Features

• Multi-camera person re-identification across heterogeneous CCTV networks
• Pattern-based behavioural and gait analysis for anomaly detection
• MPEG Media Orchestration (ISO/IEC 23090-8) for standardised stream coordination
• Agentic AI orchestration via NexusFlow for cross-source reasoning and event correlation
• Privacy-preserving analytics with selective data minimisation and differential privacy options
• Real-time alerting and event escalation for operations centres
• Replay, forensic analysis and timeline reconstruction for post-incident investigation
• EU AI Act-aligned audit trails paired with NexusTrust for full governance accountability

---

Application Areas

---

Technical Specification

Standards	ISO/IEC 23090-8 (MPEG Media Orchestration) for multi-source stream coordination
AI Approach	Re-identification networks, behavioural-pattern recognition, gait analysis
Compute	GPU-accelerated edge inference + central HPC processing
Integration	Native NexusFlow (orchestration) and NexusTrust (assurance) hooks
Privacy	Differential privacy options, on-premise deployment, GDPR-aligned
Outputs	Event alerts, person tracks, behavioural analytics, forensic timelines
Cluster	Cluster 3 — Civil Security for Society

---

How It Works

MPEG Media Orchestration

Traditional multi-camera surveillance systems treat each feed independently — operators must mentally correlate appearances across screens. NexusGuard replaces this with standardised stream orchestration following ISO/IEC 23090-8. The orchestration layer coordinates heterogeneous sources — fixed CCTV, PTZ cameras, body-worn devices, drone feeds — into a unified analytical pipeline. Each stream is registered with its spatial context, calibration parameters, and operational metadata, allowing the AI layer to reason across all sources simultaneously rather than analysing each in isolation.

Person Re-Identification

When a person of interest is identified in one camera view, NexusGuard’s re-identification networks search for matching appearances across the entire camera network — even when the person changes direction, passes through areas with different lighting, or is partially occluded. The system builds a spatio-temporal track that shows where the individual has been, when, and the confidence level of each match. This is not facial recognition alone; it combines appearance features, gait signatures, and contextual cues for robust matching across challenging real-world conditions.

Behavioural Pattern Analysis

Beyond tracking individuals, NexusGuard analyses behavioural patterns across crowds and environments. It detects anomalies — unusual loitering, counter-flow movement, abandoned objects, sudden crowd dispersal — and generates real-time alerts for operations centre staff. These alerts are accompanied by contextual evidence: the source feeds, the behavioural pattern detected, the confidence score, and recommended response actions. Operators make the decisions; the AI provides timely, structured intelligence to support them.

Forensic Timeline Reconstruction

After an incident, NexusGuard enables investigators to reconstruct a complete timeline — tracing a person’s movements across the camera network, correlating with sensor events and access logs, and producing a structured evidence pack with timestamped frames, track visualisations, and metadata provenance. Every analytical step is logged for audit, supporting evidentiary requirements and judicial review.

---

Privacy & Governance by Design

NexusGuard is designed for deployment in environments where civil liberties, data protection, and proportionality are non-negotiable. The architecture supports selective data minimisation — processing video streams for pattern detection without retaining identifiable imagery unless an alert threshold is crossed. Differential privacy techniques can be applied to aggregate analytics. All processing runs on-premise with no data leaving the deployment, and every analytical action is recorded in an immutable audit trail accessible through NexusTrust.

This governance-first approach aligns with the EU AI Act’s requirements for high-risk AI systems in law enforcement and public safety — particularly the obligations around transparency, human oversight, accuracy documentation, and risk management that apply to biometric and surveillance-adjacent technologies.

---

Heritage & Lineage

NexusGuard builds on VTG’s research lineage in agentic AI for security applications — including contributions to LASIE (forensic analysis across heterogeneous sources) and MAGNETO (knowledge representation and augmented intelligence for law enforcement) — and on standardised media-orchestration approaches to multi-source video analytics. The platform represents the convergence of VTG’s computer vision, knowledge graph, and agentic AI capabilities into a governance-ready security product.

Target Calls: Horizon Europe Cluster 3 (Civil Security for Society) — increased security, disaster-resilient society, infrastructure protection, fight against crime and terrorism, border management.